Article
Computer Science, Artificial Intelligence
Hui Wang, Zichen Wei, Gan Yu, Shuai Wang, Jiali Wu, Jiawen Liu
Summary: This paper proposes a two-stage many-objective evolutionary algorithm, TS-DGPD, which accelerates convergence and maintains population diversity by using cosine distance and Lp norm, and increases selection pressure using dynamic generalized Pareto dominance. Experimental results show that the algorithm performs well in terms of convergence and diversity.
INTERNATIONAL JOURNAL OF INTELLIGENT SYSTEMS
(2022)
Article
Computer Science, Information Systems
Jiangtao Shen, Peng Wang, Huachao Dong, Jinglu Li, Wenxin Wang
Summary: In this study, a multistage many-objective evolutionary algorithm (MaOEA) is proposed to balance convergence and diversity. Convergence and diversity are processed separately at different optimization stages, with diversity emphasized by introducing a decision variable clustering method.
INFORMATION SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Xujian Wang, Fenggan Zhang, Minli Yao
Summary: To handle the inconsistency between reference vectors (RVs) distribution and Pareto front shape in decomposition based multi-objective evolutionary algorithms, researchers have proposed various methods to adjust RVs during the evolutionary process. However, most existing algorithms adjust RVs either in each generation or at a fixed frequency without considering the evolving information of the population. To tackle this issue, the proposed MBRA algorithm adjusts RVs periodically and conditionally based on the improvement rate of convergence degree of subproblems computed through d(1) distance. Extensive experiments validate the effectiveness and competitiveness of MBRA on many-objective optimization problems, especially those with irregular Pareto fronts.
COMPLEX & INTELLIGENT SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Jiawei Yuan, Hai-Lin Liu, Fangqing Gu, Qingfu Zhang, Zhaoshui He
Summary: This article investigates the properties of ratio and difference-based indicators under the Minkovsky distance, and proposes an algorithm for solution evaluation using a ratio-based indicator. By identifying promising regions and ensuring population diversity, the algorithm demonstrates competitive performance on various benchmark problems.
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2021)
Article
Computer Science, Information Systems
Yani Xue, Miqing Li, Xiaohui Liu
Summary: This paper addresses the issue of effectiveness and efficiency in multiobjective optimization problems by proposing a novel evolutionary algorithm, E3A, which outperforms existing algorithms in quickly finding well-converged and well-diversified solutions in experiments.
INFORMATION SCIENCES
(2022)
Article
Automation & Control Systems
Kai Zhang, Gary G. Yen, Zhenan He
Summary: In this article, a recursive evolutionary algorithm EvoKnee(R) is proposed to directly search for global knee solutions and multiple local knee solutions using the minimum Manhattan distance approach, instead of a large number of Pareto optimal solutions. Unlike traditional approaches, only nondominated solutions in rank one are preserved in each generation, reducing computational cost and allowing quick convergence to knee solutions.
IEEE TRANSACTIONS ON CYBERNETICS
(2021)
Article
Computer Science, Artificial Intelligence
Zhenshou Song, Handing Wang, Cheng He, Yaochu Jin
Summary: The proposed algorithm uses Kriging-assisted two-archive EA for expensive many-objective optimization, employing an influential point-insensitive model to approximate each objective function and proposing an adaptive infill criterion for determining an appropriate sampling strategy. Experimental results have shown its superiority over five state-of-the-art SAEAs on a set of expensive multi/many-objective test problems.
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2021)
Article
Computer Science, Artificial Intelligence
Ying Xu, Huan Zhang, Xiangxiang Zeng, Yusuke Nojima
Summary: This article addresses the challenge of balancing convergence and diversity in evolutionary computation for solving many-objective optimization problems. A new preferred solution selection strategy is proposed to enhance convergence pressure by selecting non-dominated solutions with better convergence. The number of special solutions is adaptively updated to dynamically adjust the convergence pressure of the population. Experimental results demonstrate that the proposed convergence enhanced evolutionary algorithm (CEEA) outperforms state-of-the-art many-objective evolutionary algorithms.
SWARM AND EVOLUTIONARY COMPUTATION
(2022)
Article
Computer Science, Artificial Intelligence
Parviz Mohammad Zadeh, Mostafa Mohagheghi
Summary: This paper presents a novel decomposition-based hybrid many-objective optimization method using PSO and SQP algorithms. The proposed method decomposes the objective function space into optimization sub-problems and generates reference directions using the scaled objective function space and uniformly distributed points. The optimization of all the reference directions simultaneously enhances the accuracy and computational efficiency. A new mutation operator is introduced to improve the computational performance and prevent local convergence. Experimental results demonstrate that the proposed method outperforms other methods in solving many-objective optimization problems.
APPLIED INTELLIGENCE
(2023)
Article
Computer Science, Information Systems
Xujian Wang, Fenggan Zhang, Minli Yao
Summary: Research has shown that considering the convexity-concavity of Pareto fronts can improve the performance of multiobjective evolutionary algorithms (MOEAs). Based on this advantage, we propose a many-objective evolutionary algorithm, MaOEA-3C, which estimates the convexity-concavity of the Pareto fronts and incorporates clustering. The algorithm updates an elitist archive using non-dominated sorting and a niche-based method to estimate the convexity-concavity of the Pareto fronts and guides the evolving directions of the current population using clustering. The performance of MaOEA-3C is compared with seven state-of-the-art algorithms and demonstrates its effectiveness and competitiveness in many-objective optimization problems.
INFORMATION SCIENCES
(2023)
Article
Mathematics, Interdisciplinary Applications
Wen Zhong, Jian Xiong, Anping Lin, Lining Xing, Feilong Chen, Yingwu Chen
Summary: The study introduces a flexible ensemble framework ASES that enhances the performance of solving multi-objective optimization problems by embedding different MOEAs. By recording large-scale nondominated solutions in a big archive and developing an efficient strategy to update the archive, the efficiency of the algorithm is improved.
Article
Automation & Control Systems
Ying Xu, Chong Xu, Huan Zhang, Lei Huang, Yiping Liu, Yusuke Nojima, Xiangxiang Zeng
Summary: This article proposes a new metric to calculate the contribution of each decision variable to the optimization objectives, and based on this, a multiobjective evolutionary algorithm called DVCOEA is introduced. The experimental results show that DVCOEA is a competitive approach for solving large-scale multi/many-objective problems.
IEEE TRANSACTIONS ON CYBERNETICS
(2023)
Article
Automation & Control Systems
Javier Moreno, Daniel Rodriguez, Antonio J. Nebro, Jose A. Lozano
Summary: This article introduces a new efficient algorithm MNDS for computing the nondominated sorting procedure, which outperforms other techniques in terms of computational complexity and running time. The algorithm is based on the computation of the dominance set and takes advantage of the characteristics of the merge sort algorithm.
IEEE TRANSACTIONS ON CYBERNETICS
(2021)
Article
Computer Science, Artificial Intelligence
Sukrit Mittal, Dhish Kumar Saxena, Kalyanmoy Deb, Erik D. Goodman
Summary: This article presents an approach that uses machine learning to learn the relationships between top solutions in optimization problems, helping offspring solutions progress. The method involves balancing tradeoffs between convergence and diversity, using the Random Forest method, and changing the application of machine learning models.
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2022)
Article
Operations Research & Management Science
Christian von Lucken, Carlos A. Brizuela, Benjamin Baran
Summary: This work presents a new multipopulation framework for the multiobjective evolutionary algorithm based on decomposition. Clustering methods are used to reinforce mating restrictions by splitting the population into multiple subpopulations for independent evolution. The results show the viability of the clustering-based multipopulation approach in enhancing the performance of evolutionary methods for many-objective problems.
COMPUTATIONAL OPTIMIZATION AND APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Nick Zhang, Abhishek Gupta, Zefeng Chen, Yew-Soon Ong
Summary: This article introduces a novel algorithm-centric solution using evolutionary multitasking to speed up decision-making in the machine learning pipeline. By creating small data proxies and combining them with the main task, the efficiency of evolutionary search can be improved, accelerating the decision-making process. Experiments show that multitasking can significantly speed up the baseline evolutionary algorithms.
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2022)
Article
Engineering, Civil
Liang Feng, Yuxiao Huang, Ivor W. Tsang, Abhishek Gupta, Ke Tang, Kay Chen Tan, Yew-Soon Ong
Summary: The Vehicle Routing Problem is a challenging optimization problem, and this article proposes a method to speed up the optimization process by transferring knowledge from past solved problems.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Abhishek Gupta, Lei Zhou, Yew-Soon Ong, Zefeng Chen, Yaqing Hou
Summary: Evolutionary multitasking (EMT) is a concept that fills the potential gap of skill transfer between distinct optimization problems in evolutionary computation, by utilizing a population's implicit parallelism to jointly solve a set of tasks. This paper reviews various application-oriented explorations of EMT and provides recipes on how general problem formulations can be transformed in the light of EMT.
IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE
(2022)
Editorial Material
Computer Science, Artificial Intelligence
Abhishek Gupta, Yew-Soon Ong, Kenneth A. De Jong, Mengjie Zhang
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2022)
Article
Computer Science, Artificial Intelligence
Edward K. Y. Yapp, Abhishek Gupta, Xiang Li
Summary: A layer-wise neural network architecture is proposed for classification and regression of time series data with single-output instances. The approach is benchmarked against other methods and an ablation study is conducted to understand the critical design choices. The results show that the proposed method outperforms others and the parameter sharing in dense layers is key to improving performance.
JOURNAL OF INTELLIGENT MANUFACTURING
(2023)
Article
Automation & Control Systems
Lu Bai, Wu Lin, Abhishek Gupta, Yew-Soon Ong
Summary: This paper introduces a new multitasking algorithm, MTGD, and its derivative, MTESs, demonstrating faster convergence compared to the single task scenario. The theoretical findings are supported by numerical experiments on synthetic benchmarks and practical optimization examples.
IEEE TRANSACTIONS ON CYBERNETICS
(2022)
Article
Automation & Control Systems
Xinghua Qu, Yew-Soon Ong, Abhishek Gupta
Summary: This article showcases the exploration of transferability across frames to boost the creation of minimal yet powerful attacks in image-based reinforcement learning. By introducing three types of frame-correlation transfers (FCTs), the study demonstrates the tradeoff between complexity and potency of transfer mechanism, significantly speeding up attack generation on four state-of-the-art policies across six Atari games.
IEEE TRANSACTIONS ON CYBERNETICS
(2022)
Article
Multidisciplinary Sciences
Chin Sheng Tan, Abhishek Gupta, Yew-Soon Ong, Mahardhika Pratama, Puay Siew Tan, Siew Kei Lam
Summary: In multi-objective optimization, covering the Pareto front becomes challenging due to the exponential increase in points with the dimensionality of the objective space. Pareto estimation (PE) aims to overcome insufficient PF representations by using inverse machine learning. However, the accuracy of the inverse model is limited by the scarce training data in high-dimensional/expensive objectives. To address this challenge, this paper proposes multi-source inverse transfer learning for PE to enhance PF approximation. Experimental results demonstrate significant improvements in the predictive accuracy and PF approximation capacity of Pareto set learning. It envisions a future of on-demand human-machine interaction for facilitating multi-objective decisions.
SCIENTIFIC REPORTS
(2023)
Article
Computer Science, Artificial Intelligence
Han Xiang Choong, Yew-Soon Ong, Abhishek Gupta, Caishun Chen, Ray Lim
Summary: In the field of deep learning, the size of neural networks is crucial. Large pre-trained models, capable of handling various tasks and trained on extensive data, are at the forefront of artificial intelligence. However, the real-world utility of these singular models, known as Jacks of All Trades (JATs), may be limited due to resource constraints, changing objectives, and diverse task requirements. This paper explores the concept of creating a diverse set of compact machine learning models, called the Set of Sets, to address these limitations. A novel approach using a neuroevolutionary multitasking algorithm is presented, bringing us closer to collectively achieving models that are Masters of All Trades.
IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE
(2023)
Article
Computer Science, Artificial Intelligence
Xinghua Qu, Abhishek Gupta, Yew-Soon Ong, Zhu Sun
Summary: This article discusses the issue of robustness in deep reinforcement learning (DRL) policies when facing unknown perturbations. They propose an adversary agnostic robust DRL paradigm that does not require predefined adversaries. The authors provide theoretical analysis and conduct experiments to demonstrate the effectiveness of their approach.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Automation & Control Systems
Mojtaba Shakeri, Erfan Miahi, Abhishek Gupta, Yew-Soon Ong
Summary: This article proposes a novel transfer evolutionary optimization framework that enables joint evolution in the source knowledge space and the search space of solutions to the target problem, with scalability and online learning agility.
IEEE TRANSACTIONS ON CYBERNETICS
(2023)
Article
Computer Science, Artificial Intelligence
Nick Zhang, Abhishek Gupta, Zefeng Chen, Yew-Soon Ong
Summary: This study proposes a novel neuroevolutionary multitasking algorithm (NuEMT) to address the issue of high sample complexity in deep reinforcement learning. By transferring information from short-term auxiliary tasks to the target task, the algorithm enables efficient learning and evaluation of policies, reducing the requirement for expensive agent-environment interaction data.
IEEE TRANSACTIONS ON COGNITIVE AND DEVELOPMENTAL SYSTEMS
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Abhishek Gupta, Ray Lim, Chin Chun Ooi, Yew-Soon Ong
Summary: By incorporating knowledge transfer into black-box optimization with Gaussian process surrogates, the cumulative regret bounds can be tightened, leading to faster convergence and overcoming the cold start problem of traditional Bayesian optimization algorithms. Extending this method to multi-source settings further tightens the regret bounds and maintains algorithmic complexity linear in the number of sources.
2022 IEEE SYMPOSIUM SERIES ON COMPUTATIONAL INTELLIGENCE (SSCI)
(2022)
Article
Automation & Control Systems
Zefeng Chen, Abhishek Gupta, Lei Zhou, Yew-Soon Ong
Summary: In this article, a method is proposed to quickly optimize large datasets using auxiliary source tasks. A computational resource allocation strategy is designed to effectively utilize these auxiliary tasks. Experimental results show that the proposed algorithm achieves higher speedup compared to existing methods, demonstrating its efficiency in handling real-world multiobjective optimization problems involving large datasets.
IEEE TRANSACTIONS ON CYBERNETICS
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Ray Lim, Abhishek Gupta, Yew-Soon Ong
Summary: This paper investigates a data-lean variant of Transfer Evolutionary Optimization (TrEO) algorithm, which utilizes source-target similarity capture and solution representation learning to improve convergence rates. Experimental results show that this data-lean approach can achieve competitive performance.
2022 IEEE CONGRESS ON EVOLUTIONARY COMPUTATION (CEC)
(2022)